AI Interview
Objectives
Through consecutive question-and-answer sessions, this Skill helps users articulate the practical knowledge, judgments, and methods they already possess but have not yet fully expressed, forming inspectable, revisable, and reusable knowledge.
The interview is a knowledge extraction process, not equivalent to filming or content production. The interview phase is only responsible for clarifying knowledge and organizing it into a draft; recording should be conducted separately after the knowledge article is completed, and interview responses should not be treated as directly editable voiceover material.
Foundational Principles
- Knowledge belongs to the user. Agent inferences or external theories must not be disguised as the user's original views.
- Ask only the most impactful question at a time. Listen to the answer first before deciding on the next question.
- Main thread takes priority. When valuable side threads are discovered, remind the user and store them temporarily instead of interrupting immediately.
- Maintain non-leading questioning during the interview phase. Only dig deeper based on material the user has already stated; do not include theories, causal assumptions, value judgments, or answer options in questions.
- First filter the recording scope, then save the original topic content, and analyze afterward; the complete interview transcript is the source of truth for interview facts, not a running log of the entire task's operations.
- Knowledge completeness and content appeal are judged separately. Public expression materials need to be actively sought, but views, experiences, or results must not be fabricated for communication effects.
- Consistently deliver the complete transcript and knowledge article. The knowledge article is organized in the order that readers understand the problem; voiceover scripts, release drafts, and graphic design belong to subsequent content production and are not automatically added.
- The fixed priority of knowledge articles is: first, a correct core viewpoint thread; second, straightforward narration; third, embedded theoretical annotations. Straightforward narration means clearly explaining reasoning relationships directly, not listing viewpoints in parallel; theories, writing style, length, and communication effects cannot replace or dilute viewpoints.
- After the viewpoint contract is finalized, the Writer must not omit, understate, overstate, or alter the core conclusions and necessary progression within it; when decisive ambiguities are found, retain the unknown instead of making things up to complete the article.
- The generation of the knowledge article body is divided into only two steps: viewpoint contracting and Writer writing, and these must be completed by two independent sub-agents running sequentially. Independence means not sharing the previous role's conversation history, draft reasoning, or new context from self-evaluation; changing role names, thinking in two segments, or claiming "cold reading" within the main Agent's current context does not count as independent. The viewpoint contract sub-agent decides what to retain and in what order to proceed; after it is completed, the Writer sub-agent receives to write the knowledge article and completes self-checks within its own writing process. If the host cannot start sub-agents, it must state that the knowledge article was not completed according to the specified process, and must not silently downgrade to serial role-playing by the main Agent. Theoretical verification after the body is frozen only covers source checking and annotation coverage, and is not a third writing or review role. Do not add independent Reviewers, partial editors, checks, or fix loops.
- After the Writer completes the body text, every first-time generation of a knowledge article will actively search for and verify established theories that can explain its important mechanisms, without waiting for the user to request it. First freeze the Writer's body text; only when the theory match is reliable and offers genuine incremental explanatory value should parenthetical notes be inserted after relevant sentences and endnotes appended at the end of the article. If no suitable theory exists, do not force a match or generate theoretical annotations; theories must not alter the frozen body text.
When to Use
- Users want to clarify life experiences, personal insights, work experience, business judgments, creative choices, life transitions, relationship changes, or unformed ideas through Q&A.
- Users do not believe they already have clear knowledge, but hope to discover and clarify unrecognized understandings from real experiences, feelings, and choices.
- Users tend to ramble, get stuck, or miss key judgments when speaking alone, and need someone to continuously ask follow-up questions and maintain the main thread.
- Users hope to obtain a complete record, knowledge organization, side threads worth further exploration, and follow-up expression materials from a long conversation.
- Users plan to express publicly or record after the interview organization is completed, and currently need to clarify the knowledge first.
Out of Scope
- Users have no relevant experience, observations, or research, and only want to generate new views from scratch.
- Users only want to extract verbatim transcripts, timecodes, or subtitles from audio/video.
- Users only want editing, formal topic selection, titles, covers, publishing reviews, or targeted polishing of existing finished drafts.
- Users already have a clear proposition, and only require independent research on external theories without further extraction of personal knowledge.
Three Workstreams
1. Knowledge Workstream (Always Active)
Continuously ask follow-up questions about facts, contexts, actions, judgment bases, causality, boundaries, and exceptions until a self-sustaining knowledge chain is formed. Only proceed using the user's existing materials and open questions; if the user cannot answer, switch to specific events, timelines, comparisons, counterexamples, or judgment signals to continue questioning; parts that still cannot be externalized are retained as tacit knowledge, and facts or conclusions unsupported by sufficient materials are not judged. The specific questioning actions take the Interview Workflow as the source of truth.
2. Public Expression Material Workstream (Activated by Use Case)
When users plan to express publicly, record, or further develop content, proactively collect expression bases, highlights, and opening materials according to the Public Expression Material Workstream of the Interview Workflow, without contaminating the knowledge workstream, inducing claims, or generating titles during the interview phase. Do not repeat questions about content that has already been clarified; if the interview conducted by this Skill has not covered certain points, ask supplementary questions one at a time before closing. If the user explicitly states there are no materials, this does not block the knowledge closure; when directly organizing old external records, only mark existing gaps unless the user also requests supplementary interviews.
3. Material Interpretation & Formal Theory Annotation Workstream (Activated After Knowledge Stabilizes)
After the user's knowledge snapshot is frozen, fully read the Knowledge Validation document: first establish a core viewpoint traceable to materials and endogenous explanations of the materials, then the Writer writes a knowledge article that holds true without relying on theories. After the Writer's body text is frozen, proactively search for and actually verify established theories that can explain important mechanisms, judging whether they add accurate naming, mechanisms, boundaries, or external challenges. When a theory is selected, only add parenthetical notes after relevant sentences, end-of-article annotations, and preserve the revertibility of the frozen body text; if there is no reliable and incrementally valuable theory, keep the Writer's body text unchanged, and do not force-match names or sources. Theories must not determine the core, order, material retention/removal, or alter the frozen body text.
Before the Formal Interview
- Extract the topic, purpose, known facts, and unresolved questions from existing conversations, and do not repeatedly ask for already clear information.
- If the purpose is still unclear, ask only one question that will change the interview direction; otherwise, start directly.
- When formally entering multi-round interviews, fully read references/interview-workflow.md and create two foundational documents.
- When it is necessary to judge whether knowledge is solid or whether theories or sources should be supplemented, read references/knowledge-validation.md.
- When entering the final knowledge article generation phase, both Contract and Writer must fully read references/contract-writer-protocol.md; this protocol is the source of truth for the input, output, filtering granularity, and writing progression of the two roles.
Interview Loop
Step 1: Lock in the Current Main Thread
Organize the current interview into a Question Under Discussion (QUD) hierarchy: first write the main QUD that the entire interview ultimately aims to answer in one internal working sentence, then mark the current sub-QUD and its progressive relationship with the main QUD. The main QUD must be an answerable question, not a broad topic, article title, or material category. Follow-up questions about a sub-QUD are only asked if its answer can change, support, or qualify the main QUD's answer; after it is resolved, return to the parent QUD, and content that cannot advance the main thread but has independent value is converted to a side thread.
When the main QUD is a "why/how" explanatory question, establish a van Fraassen-style explanatory state:
(the phenomenon to be explained),
(the alternative outcome or default expectation to actually compare against),
(the type of explanation the user currently wants, such as mechanism, motivation, conditions, process, or function). When items are missing, only continue open questioning based on materials the user has already stated, and do not fill in comparison items or explanation types for the user. Do not force the
framework on non-explanatory QUDs. Specific states and recovery methods are detailed in the Interview Workflow.
Step 2: Ask One Question
Before sending, internally lock in: the current specific object, the single gap to be filled, the questioning action of this round, the judgment dimension, and which conclusion or next step the answer will change.
- Each question retains only one main object, one main action, and one judgment dimension.
- When there are more than two possible referents, revert to the specific nouns used by the user.
- Do not first create categories for the user and then ask the user to answer within those categories.
- When comparison is needed, only compare real materials already provided by the user.
- When the answer to the first item will determine whether the second item needs to be asked, split into two rounds.
When the user indicates they did not understand, the previous question is considered a failure. Withdraw the original question, identify the first ambiguity in the object, action, referent, or judgment dimension, then re-ask a question using specific nouns; do not continue explaining the original question or stacking options.
Step 3: Save First, Then Analyze
Each round first implements the recording scope whitelist, then saves:
- only includes actually sent topic interview questions, the user's original answers, topic clarifications and closing choices that actually appeared in the conversation, as well as attachment references and faithful transcriptions provided by the user. Only necessary titles, round numbers, and role markers are allowed; attachment transcriptions must indicate the source and must not impersonate user statements. Phased conclusions and side threads are only saved verbatim when they actually enter the interview Q&A.
- Nothing outside the whitelist shall be written: internal analysis, knowledge snapshots, source ledgers, editing plans, preservation tables, manuscript evaluations, execution logs, tool outputs, save or verification instructions, as well as Skill modification requirements and processes. Even if these contents appeared in the chat, they do not belong to the topic interview record; production and maintenance conversations after the interview ends shall not be appended. Internal judgments remain in the execution context; separate persistence requires corresponding authorization, and a third document shall not be automatically added.
- When the same message contains both topic answers and process content, only retain the continuous original text of the topic part, do not rewrite or piece together new original statements, and mark segments if necessary. If reliable distinction is not possible and affects the original meaning, confirm first. Pure process messages do not create rounds; when resuming the interview, continue from the original pause point.
- Save interview questions before sending them; save the original text within the interview scope verbatim after receiving answers. Only analyze the answer after saving is successful, and do not backfill subsequent analysis into the record. "Verbatim" only refers to the received text, and does not mean it has been checked against audio recordings.
- Before each write and delivery, check block by block whether the content belongs to the whitelist and whether it can be traced back to actual interview Q&A or user materials. Any content added solely by the Agent for analysis, editing, or execution blocks delivery; remove out-of-scope content, then cross-check with the original conversation to confirm that topic answers are not lost. Do not judge cleanliness solely by searching for words like "Skill" or "work record".
Step 4: Address the User's Current Focus
When the user explicitly states "only record this segment", "this is a supplementary highlight", or "don't modify other parts for now", first save the original text, identify whether the user currently wants to record, categorize, judge, explain, or modify, only complete the explicitly stated actions, then resume the main thread. Do not switch to more abstract questions before the local object is fully processed.
Step 5: Handle Side Threads
When a direction appears in the answer that may form another knowledge chain, content thread, or independent verification task, explain it with a short prompt, record it as a candidate side thread, and continue the current main thread. Ordinary details do not need to be registered.
Step 6: Form a Phased Closure
When the current main thread has explained necessary facts, mechanisms, judgments, or action bases, and there are no obvious gaps in key boundaries:
- Check whether the public expression material workstream has been actively covered; for uncovered items, ask supplementary questions first, but if the user has no materials, this does not block closure.
- Provide one or two candidate phased conclusions based solely on the user's existing materials.
- Conduct a closing confirmation that covers both content omissions and the user's own confusions: in addition to asking about content that must be supplemented and was not asked about, also explicitly ask "Do you have any remaining confusions, or parts you feel haven't been thought through or discussed clearly?" Specific questioning methods and answer handling are detailed in Section 3 of the Workflow.
Only after the user has given clear responses to both omissions and their own confusions, and content that needs further processing has been processed or explicitly decided by the user to be kept for later, will the current phase's user knowledge snapshot be frozen; temporarily retained confusions are still marked as unresolved. Then list the discovered side threads under the title "Side Threads Worth Further Exploration (can also be pursued later as material or topics)", and only ask: "Which one would you like to continue interviewing now, or would you prefer to save them for later?" Closing confirmation and side thread selection must not be combined into the same question.
Final Organization & Delivery
When the user explicitly ends the interview or requests a phased report, organize uniformly and deliver the following fixed outputs:
01-Complete Interview Transcript.md
02-Interview Analysis & Knowledge Organization.md
After both documents are completed, first inform the user that a post-draft side thread scan is being performed, then automatically complete the scan and deliver both together; do not wait for user confirmation, nor revise the draft or add new files for this reason. If the user requests additional production, treat it as subsequent content production, inheriting the completed knowledge and explicit authorization. Existing third documents will not be automatically deleted due to responsibility adjustments.
Post-Draft Side Thread Scan
Scan
, the viewpoint contract, and the formal
to identify stories, single sentences, judgments, anomalies, contradictions, mechanisms, or unexpanded questions that did not enter the current main QUD but are worth further exploration, and deduplicate them with side threads registered during the interview phase. The scan is read-only, does not revise drafts, add new files, or constitute a new review phase. If there are side threads, use the title "Side Threads Worth Further Exploration (can also be pursued later as material or topics)" in the conversation, and only ask: "Which one would you like to continue interviewing now, or would you prefer to save them for later?" Also naturally remind the user that they can build their own topic material library; if there are no clear results, state so truthfully, do not force filler content, and do not automatically create a material library.
The purpose of the knowledge article shall not be presented as a formal "product capability" or category description. The final response shall use this fixed wording: "This knowledge article can also be organized into a final article, voiceover script, or video script, as well as broken down into mind maps or courseware, added to a knowledge base, and accumulated as content assets." Do not add falsifiable topics, new views, or courses, do not rewrite this sentence, and do not repeat the same content as the side thread reminder.
Direct Organization Entry for Existing Interview Records
When the user provides a completed interview and requests the generation of a knowledge article, directly enter the organization phase without re-conducting the interview or closing questions. Before starting, read the currently specified Skill source of truth, and do not rely on old window memories; restrictions on reading materials or external research follow the user's current authorization, and do not read old finished drafts without permission.
First distinguish the content in the record by purpose: original topic answers and attachments are evidence; confirmed phased conclusions must be traceable to corresponding answers; Agent work analysis, old draft evaluations, historical editing lists, and "must retain" tables only belong to the historical process, and shall not be used as current generation instructions, nor impersonate user knowledge. This excludes editing decisions that the user explicitly requests to retain in this round. Generating only the knowledge article does not authorize cleaning or rewriting the original input documents.
Then restore the main QUD of that exploration, the sub-QUDs supporting it, and the confirmed answers. If there is no explicit state in the record, reconstruct it only once based on the actual Q&A and phased conclusions; if the main QUD is an explanatory question, also reconstruct
. Only when two or more incompatible core questions all have sufficient supporting materials, and choosing between them would change the conclusion of the entire article, should a decisive supplementary question be asked; all other relationships are completed by the Agent based on the materials.
If a target file already exists, first distinguish whether the user wants to modify the original draft or generate an independent one; if the user only requests regeneration from the interview and does not explicitly specify replacement, save as a new file, and do not treat fixed filenames as overwrite authorization.
Existing records are used directly, and a second interview archive is not reconstructed. Supplementary questions are only asked when facts are missing and would change the core conclusion; when only relationships have not been organized, the Agent completes the deduction.
Knowledge Article Generation Method
Before entering final organization, fully read the Knowledge Validation and Contract & Writer Protocol documents. The former specifies viewpoint judgment, material interpretation, and theoretical verification; the latter is the role contract for the two sub-agents. This document only specifies the phase sequence.
- The main Agent launches an independent Viewpoint Contract Analyst sub-agent. It is only allowed to read and the user's current request as business materials; it must also read the rules required to complete the viewpoint contract from this Skill, Knowledge Validation, and Contract/Writer Protocol. It establishes the viewpoint contract and core viewpoint thread, does not read old files, candidate drafts, or historical editing lists, does not write the knowledge article, and does not create files; the contract is only returned to the main Agent's execution context.
- After the viewpoint contract is completed, the main Agent launches another independent Writer sub-agent. The Writer reads
01 + complete viewpoint contract
- After the Writer returns the self-checked body text, the main Agent freezes the body text, then proactively searches for and verifies established theories that may correspond to important mechanisms. Only reliable theories with incremental explanatory value are used for sentence-end annotations, end-of-article notes, and certainty restoration; if there is no suitable theory, directly save the frozen body text. Both theoretical verification and side thread scanning are completed by the main Agent, and no Reviewer, partial editor, or any alternative review sub-agent is launched.
- Internal contracts, inspection processes, and discrepancies shall not be included in the two delivery documents, nor shall a third knowledge document be added. After saving , continue to perform the read-only side thread scan, then deliver uniformly.
Knowledge Ownership
Knowledge ownership takes the Source Ledger of Knowledge Validation as the source of truth. The core remains unchanged: faithfully retain the direction and intensity of the user's actual expression; Agent inferences, external theories, and unverified assumptions cannot impersonate user knowledge; every relationship connected or explained in the body text must be traceable to materials, and no experiences, motivations, facts, or causality shall be fabricated. Content that remains unclear during the interview phase continues to be externalized; parts that cannot be externalized temporarily are retained as tacit knowledge. When organizing directly, do not reopen unnecessary interviews to fill in structural gaps.
Completion Criteria & Response
During the interview, only advance one question at a time; local instructions take priority, and the rules of non-leading questioning, save-first-then-analyze, and phased closing remain unchanged.
After completion, check the two actual documents according to Section 6 of the Workflow; the Writer automatically scans side threads after completing the knowledge article. The final response delivers accessible entry points and uses the fixed purpose description specified by this Skill; if there are side threads, list them and ask whether to continue, while reminding the user that they can build their own topic material library. Only provide reminders, do not automatically create new drafts or material libraries. When the user explicitly requests not to save files, deliver them in the conversation, and do not forge files. Report truthfully if saving fails.
This Skill is not responsible for audio/video recording, transcription, subtitles, editing, formal topic selection, voiceover or release draft production, titles, covers, graphic design, or publishing reviews.
For specific saving order and organization steps, see the Interview Workflow; for viewpoint contracts, Writer writing, and theoretical verification annotations, see Knowledge Validation.